Image reconstruction / finite element modelling
|Bearbeitung:||Prof. Dr.-Ing. Udo Nackenhorst, M.Sc. Andre Lutz|
In a first step the Medical Images are read in and a region of interest (ROI) is defined to get a reduced dataset. After this several image processing techniques are used to automatically distinguish the designated data (femoral bone in this case) from surrounding tissue. The success of this procedure is dependent on the quality of the CT dataset, but works in more than 90-95% of the cases properly, even in bad quality slices. The other slices have to be segmented manually by defining the designated area with a few points to create a cubic b-spline. With the segmentation data it is possible to create a triangle surface mesh. Due to the rough texture the segmentation data has to be smoothed. This mesh is very fine an normally contains more than 100.000 triangles. To get suitable meshes for finite element applications the mesh has to be coarsend. This is done by using the "memoryless simplification" algorithm of Lindstrom and Turk, which was especially developed for large polygonal models. The coarsened triangle mesh is now ready to be converted to an FEM capable tetrahedron mesh. Especially for the issue of calculating statically equivalent load sets for simulation of stress adaptive bone remodelling it is possible to map the CT data to the FEM model. With a database of several femora a knowledge-based segmentation process would be the next progress.
Fig. 1: Part of a CT dataset (ROI) of a male hip containing more than 400 slices.
Fig. 2: Sample slices of segmented data (black border) on CT dataset: overview and closer look at the femoral head.
Fig. 3: Recovered surface models from segmentation data: model from original data and model from smoothed data. The smooth surface mesh contains of 152848 triangles. The dislocation below the femoral head has its seeds in the CT dataset and is corrected in another step.
Fig. 4: Coarsened mesh for use in finite element application containing 19779 tetrahedra and 4296 nodes.
Fig. 5: CT data mapped on extracted finite element mesh.